OpenAI has published a field report documenting eight scientific computing projects where coding agents reduced software runtimes and modernized legacy codebases. The report covers projects in genomics, immunology, statistics, and RNA sequencing.
Five of the projects used OpenAI's Codex autonomously, while three used a combination of Codex and Anthropic's Claude Code. The work fell into three categories: packaging and build-system cleanup, performance optimization, and full language or backend ports.
One project, cyvcf2, a Python library for reading genomic variant files, had its legacy build system replaced with a modern unified process. Another, HI.SIM, a DNA-sequencing read simulator, saw two autonomous optimization passes from GPT-5.2 and GPT-5.6 that cut runtime by 31 percent across a representative benchmark.
The report acknowledges a built-in caveat: it is a vendor publishing a survey of its own product's application, based on case studies written by the contributors involved. The underlying pattern it points to is real regardless. Research software has a documented maintenance problem, with tools built for single papers by small academic teams accumulating technical debt that nobody has the budget or mandate to address.
OpenAI argues that coding agents can help pay down that debt, pointing to projects where agents handled packaging refactors, performance tuning, and language ports that would otherwise require dedicated engineering time the research teams do not have.
The report does not claim the agents worked without oversight. Brent Pedersen, the contributor behind the cyvcf2 work, noted that going fast with agents is one thing, but going far in science still needs expert guidance, understanding, taste, and care.
Sources
OpenAI report links coding agents to faster science software builds - AI News
OpenAI field report - OpenAI
